CN113128323A - Remote sensing image classification method and device based on coevolution convolutional neural network learning - Google Patents
Remote sensing image classification method and device based on coevolution convolutional neural network learning Download PDFInfo
- Publication number
- CN113128323A CN113128323A CN202010047847.0A CN202010047847A CN113128323A CN 113128323 A CN113128323 A CN 113128323A CN 202010047847 A CN202010047847 A CN 202010047847A CN 113128323 A CN113128323 A CN 113128323A
- Authority
- CN
- China
- Prior art keywords
- network
- training
- parameters
- networks
- remote sensing
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Abstract
The invention discloses a remote sensing image classification method based on coevolution convolutional neural network learning, and belongs to the technical field of image processing. The method comprises the steps of initializing a plurality of same networks of different optimization methods, training the initialized networks at the same time, selecting a time interval, selecting the network with the highest classification accuracy on a test set after each time interval, storing all parameters of the model, endowing the stored model parameters to the rest networks in the collaborative training, carrying out iterative training of the network, discarding the model parameters with poor performance each time, and inheriting the model parameters with the best performance. Each optimization method can give full play to the functions, so that the optimization maximization is achieved, the convergence speed is improved, and the higher accuracy is achieved. The method can be used for scene classification of large-scale high-resolution remote sensing images, greatly improves the accuracy of remote sensing image classification, can be used in the fields of detection and evaluation of natural disasters, environmental monitoring and the like, and reduces judgment and decision errors and loss.
Description
Technical Field
The invention belongs to the technical field of image processing, and particularly relates to a remote sensing image scene classification method and device based on coevolution convolutional neural network learning.
Background
With the development and extension of deep learning, the deep learning network has made certain progress in various fields, and the remote sensing field is no exception. In recent years, the rapid development of the aviation and aerospace remote sensing technology is like tiger adding wings. Remote sensing image classification is regarded as an important application in the field of remote sensing, and is paid more and more attention by related professionals, and more energy is put into the remote sensing image classification. The remote sensing image classification method mainly comprises two categories, one category is a classification method adopting non-deep learning, and the other category is a classification method combining deep learning.
The basic flow of the traditional image non-deep learning classification method comprises the following steps: the method comprises four parts of image preprocessing, bottom layer feature extraction, feature coding, feature aggregation and classification by using a classifier. However, this method has a high dependency on artificially extracting features. Manually extracting features is time consuming and requires associated expertise. In the big data era, the image classification is required to be efficiently and accurately completed, and the image classification can not be completed only by manually extracting the features. At the moment, the deep learning network is a great useful place. The basic idea is to learn hierarchical feature expression in a supervised or unsupervised mode to complete feature description of the image from a bottom layer to a high layer. The deep learning network can learn strong feature expression by autonomously extracting image features in a layer-by-layer training mode. Therefore, the classification method based on the deep learning network can achieve a good effect when the remote sensing image scene is classified. The existing deep learning network is mainly established on the basis of a convolutional neural network, and the AlexNet, the VGGNet and the deep residual error network which are widely applied are obtained.
Compared with a common image data set, the remote sensing image data set has the following problems:
(1) the data scale is small: the existing remote sensing data set has fewer scene categories, and each category contains fewer pictures;
(2) the data lacks diversity: because the data is small in integral scale, image information provided by the data set is naturally not rich enough.
The development of deep learning networks in scene classification applications is greatly limited by the problems of remote sensing data sets. The data size is limited, resulting in the fact that these network-learned feature expressions are not robust, and the network is prone to overfitting situations.
Disclosure of Invention
The purpose of the invention is as follows: aiming at the problems, the invention provides a remote sensing image classification method based on coevolution convolutional neural network learning. Compared with the existing remote sensing image classification technology, the method is beneficial to improving the accuracy of remote sensing image scene classification. In addition, the remote sensing image classification method based on the coevolution convolutional neural network learning adopts coevolution in the training process, and fully utilizes the advantages of each network, so that the network can be converged more quickly and higher classification accuracy is achieved.
The technical scheme is as follows: in order to realize the purpose of the invention, the technical scheme adopted by the invention is as follows: a remote sensing image classification method based on coevolution convolution neural network learning comprises the following steps:
firstly, initializing a plurality of identical networks based on different optimization methods, training a plurality of neural networks simultaneously in a training stage, obtaining the accuracy of each network in a testing stage, finding out the network with the best performance, discarding the network parameters of the rest networks in the collaborative training, inheriting all the network parameters of the network with the highest accuracy, and carrying out iterative training according to the collaborative evolution until the set training termination times are reached. The method specifically comprises the following steps:
s1, constructing a training data set and a test set;
s1.1, dividing an original remote sensing image data set into a training set and a testing set, and processing pictures of the training set and the testing set;
s2 initializes the three networks;
s2.1, setting a network model structure used for training, wherein the models selected by the three networks are the same, and initializing the models;
s2.2, selecting appropriate and same loss functions for the three networks;
s2.3, selecting different optimization methods of three network parameters, respectively endowing the selected optimization methods to the three networks after the network model is initialized, and initializing the network parameter optimization method for each network;
s3 training the initialized three networks;
s3.1, training the three networks simultaneously according to the optimization method of each network, and updating the network parameters of each network respectively;
s4 testing the classification accuracy of the trained three networks;
s5, selecting the network with the highest test accuracy, and endowing all network parameters to the rest networks;
and S6 iterative training, importing the data set again, updating self parameters of each network in one training, and after one training is finished, updating parameters among the networks according to the strategy of S5 until the set training times are reached.
Further, step S1.1 is specifically as follows:
s1.1.1, dividing a remote sensing image data set, and dividing each type in the data set according to the ratio of 6: 1, respectively dividing the ratio into a training set and a test set;
s1.1.2 processing the training set and the test set, adjusting the size of the picture from 256 × 256 to 224 × 224, namely obtaining the picture by adopting a central cutting or random cutting mode, and performing data enhancement by a horizontal turning method;
step S2.1 is specifically as follows:
s2.1.1 selecting a network model with the processed pictures as input and the categories as output, and modifying the number of categories output by the fully-connected classifier according to the number of categories in the data set:
classifer＝Linear(m,n)；
wherein m represents the number of channels output by an upper network, and n represents the number of categories output by a classifier;
s2.1.2 according to the complexity of the selected network model, adjusting the structure parameters of the model to prevent the occurrence of overfitting phenomenon.
Further, the step S2.2 specifically includes the following steps:
s2.2.1 Cross _ entry is experimentally selected as a loss function of the network to measure the distance between the true value and the predicted value:
wherein p iskClass representing true value, i.e. kth class, qkRepresenting the prediction classes output by the network classifier, n representing the number of classes;
by using Cross EntropyLoss, the problem is expressed as a convex optimization problem, and the convex optimization problem has good convergence when a plurality of optimization methods are used;
step S2.3 comprises the following steps:
s2.3.1 setting the first optimization method of network minimization loss function as random gradient descent (SGD);
s2.3.2 setting the second network minimization loss function optimization method to Adam;
s2.3.3 the third optimization method for minimizing the loss function of the network is set as RMSprop.
Further: the step S3.1 comprises the following steps:
s3.1.1 sending training data to the network in batch, training all training data once to represent the completion of network training;
s3.1.2 in order to improve the training efficiency, the training of each network is used as a thread, all networks are trained simultaneously, the loss function is minimized by the optimization method of each network, and the network parameters are updated by back propagation;
the step S4 specifically includes the following steps:
s4, the network training is performed by taking one time as a basic time unit, and when all networks finish one-time training, all networks are tested on a test set for the classification accuracy rate;
the step S5 specifically includes the following steps:
s5, comparing the classification accuracy rates output by all networks, selecting the network with the most superiority, namely the network with the highest classification accuracy rate, keeping the network parameters of the network unchanged, discarding the network parameters of other networks, and inheriting all the network parameters of the network with the highest accuracy rate;
the step S6 specifically includes the following steps:
and S6, after each training is finished, the most advantageous network keeps the network parameters thereof, and the rest networks update all the network parameters, so as to be used as the initial conditions of the next training for training until the set training times are reached.
In addition, the invention also provides a remote sensing image classification device based on the coevolution convolutional neural network, which comprises:
the data construction module is used for dividing a training set and a test set of remote sensing image data and preprocessing pictures;
the network initial module is used for selecting a network model for realizing remote sensing image classification, applying the network model to all networks and setting different optimization methods for each network;
the network training module is used for minimizing a loss function and updating network parameters by the network according to respective optimization methods;
the testing module tests after each training is finished and selects the network with the highest classification accuracy;
the network updating module is used for updating all the parameters of the rest networks into the parameters of the network with the highest classification accuracy;
and the iteration module is used for performing iterative training, importing the data set again, updating the parameters of each network in one training, and after one training is completed, updating the parameters among the networks according to the strategy of the network updating module until the set training times are reached.
Further, the data construction module: dividing a remote sensing image data set, and enabling each type in the data set to be 6: 1, respectively dividing the ratio into a training set and a test set; processing the training set and the test set, adjusting the size of the picture from 256 to 224, namely acquiring the picture by adopting a central cutting or random cutting mode, and performing data enhancement by a horizontal turning method;
the network initialization module:
setting a model: setting three network model structures used for training, wherein the models selected by the three networks are the same, and initializing the models;
selecting a loss function: selecting appropriate and identical loss functions for the three networks;
setting different parameter optimization methods: selecting different optimization methods of three network parameters, respectively endowing the selected optimization methods to the three network models after the network models are initialized, and initializing the network parameter optimization method for each network;
the network training module:
allocating threads: distributing a thread for each network, wherein the granularity of the thread is a batch of data to be tested, and each batch of data is a preset number of pictures;
thread training: the network is trained in the respective assigned threads, and the loss function is minimized by using the optimizer, so that the parameters of the network are updated;
the test module is used for:
allocating threads: distributing a thread for each trained network model, wherein the granularity of the thread is a batch of test data, and each batch of data is a preset number of pictures;
and (3) thread testing: testing the trained network models in the respective assigned threads to obtain the number of pictures classified correctly in a batch of data;
the network update module:
find the most advantageous network: comparing the accuracy of all the networks in the test to obtain the network with the highest classification accuracy;
changing network parameters: except that the network parameters of the network with the highest accuracy rate are kept unchanged, the other networks discard the network parameters of the other networks and inherit the network parameters of the network with the highest accuracy rate.
Has the advantages that: compared with the prior art, the technical scheme of the invention has the following beneficial technical effects:
firstly, because the invention adopts a multi-network training method, by training the networks of different optimization methods and continuously updating parameters among the networks based on co-evolution, the characteristic advantages of each network, namely the advantages of each optimization method, are fully exerted, the convergence speed of the networks is improved, and the classification accuracy is improved;
second, the present invention starts a search from a set of strings of problem solutions based on co-evolution, rather than starting from a single solution. The coevolution training method starts to search from the cluster set, has large coverage and is beneficial to global preference;
thirdly, in the process of training a plurality of networks, a multi-thread operation mode is adopted, namely each network corresponds to one thread, and the networks are trained in each thread simultaneously, so that the training concurrency is improved, and the training time of the plurality of networks is greatly shortened.
Drawings
FIG. 1 is a diagram of steps of a remote sensing image classification method based on co-evolution convolutional neural network learning according to the present invention;
FIG. 2 is a block diagram of a remote sensing image classification device based on co-evolution convolutional neural network learning according to a first embodiment of the present invention.
Detailed Description
The technical scheme of the invention is further explained in detail by combining the drawings in the specification.
Referring to fig. 1, the method of the invention includes the following steps: a remote sensing image classification method based on coevolution convolution neural network learning comprises the following steps:
firstly, initializing a plurality of identical networks based on different optimization methods, training a plurality of neural networks simultaneously in a training stage, obtaining the accuracy of each network in a testing stage, finding out the network with the best performance, discarding the network parameters of the rest networks in the collaborative training, inheriting all the network parameters of the network with the highest accuracy, and carrying out iterative training according to the collaborative evolution until the set training termination times are reached. The method specifically comprises the following steps:
s1, constructing a training data set and a test set;
s1.1, dividing an original remote sensing image data set into a training set and a testing set, and processing pictures of the training set and the testing set;
s2 initializes the three networks;
s2.1, setting a network model structure used for training, wherein the models selected by the three networks are the same, and initializing the models;
s2.2, selecting appropriate and same loss functions for the three networks;
s2.3, selecting different optimization methods of three network parameters, respectively endowing the selected optimization methods to the three networks after the network model is initialized, and initializing the network parameter optimization method for each network;
s3 training the initialized three networks;
s3.1, training the three networks simultaneously according to the optimization method of each network, and updating the network parameters of each network respectively;
s4 testing the classification accuracy of the trained three networks;
s5, selecting the network with the highest test accuracy, and endowing all network parameters to the rest networks;
and S6 iterative training, importing the data set again, updating self parameters of each network in one training, and after one training is finished, updating parameters among the networks according to the strategy of S5 until the set training times are reached.
Further, step S1.1 is specifically as follows:
s1.1.1, dividing a remote sensing image data set, and dividing each type in the data set according to the ratio of 6: 1, respectively dividing the ratio into a training set and a test set;
s1.1.2 processing the training set and the test set, adjusting the size of the picture from 256 × 256 to 224 × 224, namely obtaining the picture by adopting a central cutting or random cutting mode, and performing data enhancement by a horizontal turning method;
step S2.1 is specifically as follows:
s2.1.1 selecting a network model with the processed pictures as input and the categories as output, and modifying the number of categories output by the fully-connected classifier according to the number of categories in the data set:
classifer＝Linear(m,n)；
wherein m represents the number of channels output by an upper network, and n represents the number of categories output by a classifier;
s2.1.2 according to the complexity of the selected network model, adjusting the structure parameters of the model to prevent the occurrence of overfitting phenomenon.
Further, the step S2.2 specifically includes the following steps:
s2.2.1 Cross _ entry is experimentally selected as a loss function of the network to measure the distance between the true value and the predicted value:
wherein p iskClass representing true value, i.e. kth class, qkRepresenting the prediction classes output by the network classifier, n representing the number of classes;
by using Cross EntropyLoss, the problem is expressed as a convex optimization problem, and the convex optimization problem has good convergence when a plurality of optimization methods are used;
step S2.3 comprises the following steps:
s2.3.1 setting the first optimization method of network minimization loss function as random gradient descent (SGD);
s2.3.2 setting the second network minimization loss function optimization method to Adam;
s2.3.3 the third optimization method for minimizing the loss function of the network is set as RMSprop.
Further: the step S3.1 comprises the following steps:
s3.1.1 sending training data to the network in batch, training all training data once to represent the completion of network training;
s3.1.2 in order to improve the training efficiency, the training of each network is used as a thread, all networks are trained simultaneously, the loss function is minimized by the optimization method of each network, and the network parameters are updated by back propagation;
the step S4 specifically includes the following steps:
s4, the network training is performed by taking one time as a basic time unit, and when all networks finish one-time training, all networks are tested on a test set for the classification accuracy rate;
the step S5 specifically includes the following steps:
s5, comparing the classification accuracy rates output by all networks, selecting the network with the most superiority, namely the network with the highest classification accuracy rate, keeping the network parameters of the network unchanged, discarding the network parameters of other networks, and inheriting all the network parameters of the network with the highest accuracy rate;
the step S6 specifically includes the following steps:
and S6, after each training is finished, the most advantageous network keeps the network parameters thereof, and the rest networks update all the network parameters, so as to be used as the initial conditions of the next training for training until the set training times are reached.
In addition, the invention also provides a remote sensing image classification device based on the coevolution convolutional neural network, which comprises:
the data construction module is used for dividing a training set and a test set of remote sensing image data and preprocessing pictures;
the network initial module is used for selecting a network model for realizing remote sensing image classification, applying the network model to all networks and setting different optimization methods for each network;
the network training module is used for minimizing a loss function and updating network parameters by the network according to respective optimization methods;
the testing module tests after each training is finished and selects the network with the highest classification accuracy;
the network updating module is used for updating all the parameters of the rest networks into the parameters of the network with the highest classification accuracy;
and the iteration module is used for performing iterative training, importing the data set again, updating the parameters of each network in one training, and after one training is completed, updating the parameters among the networks according to the strategy of the network updating module until the set training times are reached.
Further, the data construction module: dividing a remote sensing image data set, and enabling each type in the data set to be 6: 1, respectively dividing the ratio into a training set and a test set; processing the training set and the test set, adjusting the size of the picture from 256 to 224, namely acquiring the picture by adopting a central cutting or random cutting mode, and performing data enhancement by a horizontal turning method;
the network initialization module:
setting a model: setting three network model structures used for training, wherein the models selected by the three networks are the same, and initializing the models;
selecting a loss function: selecting appropriate and identical loss functions for the three networks;
setting different parameter optimization methods: selecting different optimization methods of three network parameters, respectively endowing the selected optimization methods to the three network models after the network models are initialized, and initializing the network parameter optimization method for each network;
the network training module:
allocating threads: distributing a thread for each network, wherein the granularity of the thread is a batch of data to be tested, and each batch of data is a preset number of pictures;
thread training: the network is trained in the respective assigned threads, and the loss function is minimized by using the optimizer, so that the parameters of the network are updated;
the test module is used for:
allocating threads: distributing a thread for each trained network model, wherein the granularity of the thread is a batch of test data, and each batch of data is a preset number of pictures;
and (3) thread testing: testing the trained network models in the respective assigned threads to obtain the number of pictures classified correctly in a batch of data;
the network update module:
find the most advantageous network: comparing the accuracy of all the networks in the test to obtain the network with the highest classification accuracy;
changing network parameters: except that the network parameters of the network with the highest accuracy rate are kept unchanged, the other networks discard the network parameters of the other networks and inherit the network parameters of the network with the highest accuracy rate.
The effects of the present invention are further explained below:
1. the experimental conditions are as follows:
the experiments of the invention were performed in a hardware environment of a dual NVIDIA GTX 1080Ti GPU and in a software environment of Python.
The remote sensing data set used in the experiment of the invention is NWPU-RESISC45, wherein the pseudo data sets for verifying the effectiveness of the co-evolution are CIFAR-10 and CIFAR-100.
NWPU-RESISC45 is a public remote sensing data set, released in 2016. This data set has 31500 pictures in total, 45 categories with 700 pictures per category. These 45 categories are: an airplane, an airport, a baseball field, a basketball court, a beach, a bridge, a bush, a church, a round farmland, a cloud, a commercial area, a dense residential area, a desert, a forest, a highway, a golf course, an athletic field, a harbor, an industrial area, an intersection, an island, a lake, a grassland, a medium residential area, a prefabricated house, a mountain, an overpass, a palace, a parking lot, a train station, a square farmland, a river, a roundabout, a runway, sea ice, a ship, an iceberg, a sparse residential area, a stadium, a storage tank, a tennis court, a terrace, a thermal power plant, and a wetland.
2. And (4) analyzing results:
the simulation experiment of the invention adopts the method of the invention and the unchanged VGG19 network to classify the optical image data set and the remote sensing data set, and carries out comparative analysis with the classification result trained by a single network.
The following table is a statistical table comparing the overall accuracy of the experiment of the present invention with VGG neural networks using three separate optimization methods and the method of the present invention.
In the following table, "Data Set" indicates the type of the adopted Data Set, "Class" indicates the scene type corresponding to the adopted Data Set, "Methods" indicates the type of the adopted classification method, and "Accuracy" indicates the Accuracy of the classification.
TABLE 1 CIFAR-10 image Classification result comparison List
TABLE 2 CIFAR-100 comparison List of image classification results
TABLE 3 NWPU-RESISC45 comparison List of image Classification results
As can be seen from the result table, the classification accuracy of the method is higher on the three data sets than that of the convolutional network method of network training by adopting a single optimization method.
It should be noted that, in the present specification, the embodiments are all described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments may be referred to each other. For the device-like embodiment, since it is basically similar to the method embodiment, the description is simple, and for the relevant points, reference may be made to the partial description of the method embodiment.
Finally, it should also be noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. The term "comprising", without further limitation, means that the element so defined is not excluded from the group consisting of additional identical elements in the process, method, article, or apparatus that comprises the element.
For convenience of description, the above devices are described as being divided into various units by function, and are described separately. Of course, the functionality of the units may be implemented in one or more software and/or hardware when implementing the present application.
From the above description of the embodiments, it is clear to those skilled in the art that the present application can be implemented by software plus necessary general hardware platform. Based on such understanding, the technical solutions of the present application may be essentially or partially implemented in the form of a software product, which may be stored in a storage medium, such as a ROM/RAM, a magnetic disk, an optical disk, etc., and includes several instructions for enabling a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method according to the embodiments or some parts of the embodiments of the present application.
The above detailed description is provided for a retrieval method, apparatus and system provided by the present application, and the principle and implementation of the present application are explained by applying specific examples, and the description of the above embodiments is only used to help understanding the method and core ideas of the present application; meanwhile, for a person skilled in the art, according to the idea of the present application, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present application.
Claims (6)
1. A remote sensing image classification method based on co-evolution convolutional neural network learning is characterized by comprising the following steps:
s1, constructing a training data set and a test set;
s1.1, dividing an original remote sensing image data set into a training set and a testing set, and processing pictures of the training set and the testing set;
s2 initializes three network models;
s2.1, setting three network model structures used for training, wherein the models selected by the three networks are the same, and initializing the models;
s2.2, selecting appropriate and same loss functions for the three network models;
s2.3, selecting different optimization methods of three network parameters, respectively endowing the selected optimization methods to the three network models after the network models are initialized, and initializing the network parameter optimization method for each network model;
s3 training three initialized network models;
s3.1, training the three network models simultaneously according to the optimization method of each network model, and updating the network parameters of each network respectively;
s4, testing the classification accuracy of the trained three network models by using the pictures of the test set;
s5, selecting the network model with the highest test accuracy, and endowing all network parameters of the network model with the rest network models;
and S6 iterative training, importing the data set again, performing self parameter updating on each network model in one training, and after one training is completed, performing parameter updating between the network models according to the strategy of S5 until the set training times are reached.
2. The remote sensing image classification method based on the coevolution convolutional neural network learning as claimed in claim 1, wherein the step S1.1 is as follows:
s1.1.1, dividing a remote sensing image data set, and dividing each type in the data set according to the ratio of 6: 1, respectively dividing the ratio into a training set and a test set;
s1.1.2 processing the training set and the test set, adjusting the size of the picture from 256 × 256 to 224 × 224, namely adjusting the picture by adopting a center cutting or random cutting mode, and then performing data enhancement by a horizontal turning method;
step S2.1 is specifically as follows:
s2.1.1 selecting a network model with the processed pictures as input and the categories as output, and modifying the number of categories output by the fully-connected classifier according to the number of categories in the data set:
classifer＝Linear(m,n)；
wherein m represents the number of channels output by an upper network, and n represents the number of categories output by a classifier;
s2.1.2 the structural parameters of the model are adjusted according to the complexity of the selected network model.
3. The remote sensing image classification method based on the coevolution convolutional neural network learning as claimed in claim 1, wherein the step S2.2 comprises the following steps:
s2.2.1 selects cross _ entry as the loss function of the net to measure the distance between the true value and the predicted value:
wherein p iskClass representing true value, i.e. kth class, qkRepresenting the prediction classes output by the network classifier, n representing the number of classes;
the optimization method for selecting three different network parameters in step S2.3 is as follows:
s2.3.1, setting the optimization method of the first network model minimizing loss function as a random gradient descent method (SGD);
s2.3.2 setting the optimization method of the second network model minimizing loss function as Adam;
s2.3.3 the third network model is set to the RMSprop optimization method.
4. The remote sensing image classification method based on the coevolution convolutional neural network learning of claim 1, which is characterized in that:
the step S3.1 comprises the following steps:
s3.1.1, sending the training data to the network model in batch, and training all the training data once to represent that the network model is trained once;
s3.1.2 taking the training of each network model as a thread, training all networks simultaneously, minimizing loss function by using the optimization method of each network, and updating network parameters by using back propagation;
the step S4 specifically includes the following steps:
s4, taking one-time training of the network models as a basic time unit, and testing the classification accuracy of all networks on a test set when all the network models finish one-time training;
the step S5 specifically includes the following steps:
s5, comparing the classification accuracy rates output by all networks, selecting the network with the most superiority, namely the network with the highest classification accuracy rate, keeping the network parameters of the network unchanged, discarding the network parameters of other networks, and inheriting all the network parameters of the network with the highest accuracy rate;
the step S6 specifically includes the following steps:
and S6, after each training is finished, the most advantageous network keeps the network parameters thereof, and the rest networks update all the network parameters, so as to be used as the initial conditions of the next training for training until the set training times are reached.
5. A remote sensing image classification device based on a coevolution convolution neural network is characterized by comprising the following modules:
the data construction module is used for dividing a training set and a test set of remote sensing image data and preprocessing pictures;
the network initial module is used for selecting a network model for realizing remote sensing image classification, applying the network model to all network models and setting different optimization methods for each network model;
the network training module is used for minimizing a loss function and updating network parameters by the network according to respective optimization methods;
the testing module tests after each training is finished and selects the network with the highest classification accuracy;
the network updating module is used for updating all the parameters of the rest networks into the parameters of the network with the highest classification accuracy;
and the iteration module is used for performing iterative training, importing the data set again, updating the parameters of each network in one training, and after one training is completed, updating the parameters among the networks according to the strategy of the network updating module until the set training times are reached.
6. The remote sensing image classification device based on the coevolution convolutional neural network learning as claimed in claim 5, wherein the data construction module: dividing a remote sensing image data set, and enabling each type in the data set to be 6: 1, respectively dividing the ratio into a training set and a test set; processing the training set and the test set, adjusting the size of the picture from 256 × 256 to 224 × 224, namely adjusting by adopting a central cutting or random cutting mode, and performing data enhancement by a horizontal turning method;
the network initialization module:
setting a model: setting three network model structures used for training, wherein the models selected by the three network models are the same, and initializing the models;
selecting a loss function: selecting appropriate and same loss functions for the three network models;
setting different parameter optimization methods: selecting different optimization methods of three network parameters, respectively endowing the selected optimization methods to the three network models after the network models are initialized, and initializing the network parameter optimization method for each network model;
the network training module:
allocating threads: distributing a thread for each network, wherein the granularity of the thread is a batch of data to be tested, and each batch of data is a preset number of pictures;
thread training: the network is trained in the respective assigned threads, and the loss function is minimized by using the optimizer, so that the parameters of the network are updated;
the test module is used for:
allocating threads: distributing a thread for each trained network model, wherein the granularity of the thread is a batch of test data, and each batch of data is a preset number of pictures;
and (3) thread testing: testing the trained network models in the respective assigned threads to obtain the number of pictures classified correctly in a batch of data;
the network update module:
find the most advantageous network: comparing the accuracy of all the networks in the test to obtain the network with the highest classification accuracy;
changing network parameters: except that the network parameters of the network with the highest accuracy rate are kept unchanged, the other networks discard the network parameters of the other networks and inherit the network parameters of the network with the highest accuracy rate.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010047847.0A CN113128323B (en) | 2020-01-16 | 2020-01-16 | Remote sensing image classification method and device based on co-evolution convolutional neural network learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010047847.0A CN113128323B (en) | 2020-01-16 | 2020-01-16 | Remote sensing image classification method and device based on co-evolution convolutional neural network learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113128323A true CN113128323A (en) | 2021-07-16 |
| CN113128323B CN113128323B (en) | 2023-08-18 |
Family
ID=76771767
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010047847.0A Active CN113128323B (en) | 2020-01-16 | 2020-01-16 | Remote sensing image classification method and device based on co-evolution convolutional neural network learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113128323B (en) |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108171266A (en) * | 2017-12-25 | 2018-06-15 | 中国矿业大学 | A kind of learning method of multiple target depth convolution production confrontation network model |
| CN108805200A (en) * | 2018-06-08 | 2018-11-13 | 中国矿业大学 | Optical remote sensing scene classification method and device based on the twin residual error network of depth |
| CN109583425A (en) * | 2018-12-21 | 2019-04-05 | 西安电子科技大学 | A kind of integrated recognition methods of the remote sensing images ship based on deep learning |
| CN109670528A (en) * | 2018-11-14 | 2019-04-23 | 中国矿业大学 | The data extending method for blocking strategy at random based on paired samples towards pedestrian's weight identification mission |
| CN109784376A (en) * | 2018-12-26 | 2019-05-21 | 中国科学院长春光学精密机械与物理研究所 | A kind of Classifying Method in Remote Sensing Image and categorizing system |
| WO2019139460A2 (en) * | 2019-02-01 | 2019-07-18 | 엘지전자 주식회사 | Artificial intelligence apparatus |
| CN110321811A (en) * | 2019-06-17 | 2019-10-11 | 中国工程物理研究院电子工程研究所 | Depth is against the object detection method in the unmanned plane video of intensified learning |
| CN110533097A (en) * | 2019-08-27 | 2019-12-03 | 腾讯科技（深圳）有限公司 | A kind of image definition recognition methods, device, electronic equipment and storage medium |
| US20190391666A1 (en) * | 2019-07-01 | 2019-12-26 | Lg Electronics Inc. | Gesture recognition apparatus and method |
| CN110619632A (en) * | 2019-09-18 | 2019-12-27 | 华南农业大学 | Mango example confrontation segmentation method based on Mask R-CNN |
| CN110636020A (en) * | 2019-08-05 | 2019-12-31 | 北京大学 | Neural network equalization method for adaptive communication system |
-
2020
- 2020-01-16 CN CN202010047847.0A patent/CN113128323B/en active Active
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108171266A (en) * | 2017-12-25 | 2018-06-15 | 中国矿业大学 | A kind of learning method of multiple target depth convolution production confrontation network model |
| CN108805200A (en) * | 2018-06-08 | 2018-11-13 | 中国矿业大学 | Optical remote sensing scene classification method and device based on the twin residual error network of depth |
| CN109670528A (en) * | 2018-11-14 | 2019-04-23 | 中国矿业大学 | The data extending method for blocking strategy at random based on paired samples towards pedestrian's weight identification mission |
| CN109583425A (en) * | 2018-12-21 | 2019-04-05 | 西安电子科技大学 | A kind of integrated recognition methods of the remote sensing images ship based on deep learning |
| CN109784376A (en) * | 2018-12-26 | 2019-05-21 | 中国科学院长春光学精密机械与物理研究所 | A kind of Classifying Method in Remote Sensing Image and categorizing system |
| WO2019139460A2 (en) * | 2019-02-01 | 2019-07-18 | 엘지전자 주식회사 | Artificial intelligence apparatus |
| CN110321811A (en) * | 2019-06-17 | 2019-10-11 | 中国工程物理研究院电子工程研究所 | Depth is against the object detection method in the unmanned plane video of intensified learning |
| US20190391666A1 (en) * | 2019-07-01 | 2019-12-26 | Lg Electronics Inc. | Gesture recognition apparatus and method |
| CN110636020A (en) * | 2019-08-05 | 2019-12-31 | 北京大学 | Neural network equalization method for adaptive communication system |
| CN110533097A (en) * | 2019-08-27 | 2019-12-03 | 腾讯科技（深圳）有限公司 | A kind of image definition recognition methods, device, electronic equipment and storage medium |
| CN110619632A (en) * | 2019-09-18 | 2019-12-27 | 华南农业大学 | Mango example confrontation segmentation method based on Mask R-CNN |
Non-Patent Citations (4)
| Title |
|---|
| SOMENATH BERA等: "Analysis of various optimizers on deep convolutional neural network model in the application of hyperspectral remote sensing image classification", 《INTERNATIONAL JOURNAL OF REMOTE SENSING 》, pages 2664 - 2683 * |
| WEI ZHAO等: "Superpixel-Based Multiple Local CNN for Panchromatic and Multispectral Image Classification", 《IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING》, vol. 55, no. 7, XP011654084, DOI: 10.1109/TGRS.2017.2689018 * |
| 牛奔;李丽;: "基于MCPSO算法的BP神经网络训练", 深圳大学学报(理工版), no. 02, pages 147 - 150 * |
| 谭朝文;王威廉;宗容;潘家华;朱莉莉;: "卷积神经网络应用于先心病心音信号分类研究", 计算机工程与应用, no. 12, pages 174 - 180 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113128323B (en) | 2023-08-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108805200B (en) | Optical remote sensing scene classification method and device based on depth twin residual error network | |
| Shi et al. | Domain adaption for fine-grained urban village extraction from satellite images | |
| Tang et al. | EMTCAL: Efficient multiscale transformer and cross-level attention learning for remote sensing scene classification | |
| CN109948029A (en) | Based on the adaptive depth hashing image searching method of neural network | |
| CN110334765A (en) | Remote Image Classification based on the multiple dimensioned deep learning of attention mechanism | |
| CN114092832B (en) | High-resolution remote sensing image classification method based on parallel hybrid convolutional network | |
| CN112507793A (en) | Ultra-short-term photovoltaic power prediction method | |
| CN111241306B (en) | Path planning method based on knowledge graph and pointer network | |
| CN110164129B (en) | Single-intersection multi-lane traffic flow prediction method based on GERNN | |
| Xue et al. | An efficient deep-sea debris detection method using deep neural networks | |
| CN112070729A (en) | Anchor-free remote sensing image target detection method and system based on scene enhancement | |
| Guo et al. | Network pruning for remote sensing images classification based on interpretable CNNs | |
| CN113221950A (en) | Graph clustering method and device based on self-supervision graph neural network and storage medium | |
| CN114943345A (en) | Federal learning global model training method based on active learning and model compression | |
| CN110555461A (en) | scene classification method and system based on multi-structure convolutional neural network feature fusion | |
| CN111582091A (en) | Pedestrian identification method based on multi-branch convolutional neural network | |
| Li et al. | Efficient detection in aerial images for resource-limited satellites | |
| CN114004152B (en) | Multi-wind-field wind speed space-time prediction method based on graph convolution and recurrent neural network | |
| Shi et al. | EBNAS: Efficient binary network design for image classification via neural architecture search | |
| CN113191445B (en) | Large-scale image retrieval method based on self-supervision countermeasure Hash algorithm | |
| CN111783688B (en) | Remote sensing image scene classification method based on convolutional neural network | |
| CN110490234A (en) | The construction method and classification method of classifier based on Cluster Classification associative mechanism | |
| Wan et al. | E2SCNet: Efficient multiobjective evolutionary automatic search for remote sensing image scene classification network architecture | |
| CN112446305A (en) | Pedestrian re-identification method based on classification weight equidistant distribution loss model | |
| CN113128323A (en) | Remote sensing image classification method and device based on coevolution convolutional neural network learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |